Mistral announced Large 4 on 6 October 2026 as a public preview of a very large multimodal mixture-of-experts model. Its announcement promised downloadable weights later in the month. As of the editorial cutoff of 25 October, the announcement is evidence of an API preview, not proof that the weights had already been released for self-hosting. The difference changes procurement, infrastructure design and licensing discussions.
Architecture in context
Mistral describes about one trillion total parameters with tens of billions active for each token. Mixture-of-experts means only part of the network is activated for a given token; it does not mean a self-hosted installation needs memory only for active parameters. Actual memory and throughput depend on total weights, precision, KV cache, model-parallel partitioning, batch sizes and serving implementation.
Endpoint and readiness
The Mistral changelog identifies mistral-large-4 as Public Preview from 6 October and documents a one-million-token context. Treat any promotional launch discount as historical, not a perpetual API price. A preview may evolve or be subject to different support guarantees. Record the precise endpoint and request surface, and avoid assigning a stable production SLA by implication.
Hosting reality
Self-hosting is a future option only after the model weights and binding license have actually been published. Evaluate GPU supply, network cost, observability, patches, authentication, moderation and distribution rights. A license reported for Mistral 3 does not automatically attach to Mistral Large 4. Keep license null until the Large 4 release materials can be verified.
Alternatives
Mistral Large 3 was explicitly released in December 2025 with Apache 2.0 weights. Teams requiring an inspectable local artifact may prefer a confirmed published open-weight release over a newer preview with unpublished weights. For hosted comparisons, measure successful task cost, not only headline parameters.
What to verify before adopting this model or method
Use the exact release ID and API endpoint; confirm your geographic processing location, entitlement tier, model lifecycle, feature flags, context and output limits from the provider console. Run a reproducible evaluation on your own data and retain human oversight for consequential actions. Where a specification, price, commercial license or entitlement is not verified for your deployment, this article deliberately leaves it unasserted instead of filling a plausible number.
Documentation clarification (25 October 2026)
Mistral’s October 6, 2026 changelog identifies mistral-large-4 as a Public Preview and says open weights are coming soon. The API preview does not by itself prove a downloadable checkpoint, license or production SLA.
Sources & Further Reading
- mistral.ai — documentation/reference (supplier-provided reference; not independently checked in this pass)
- docs.mistral.ai — documentation/reference (supplier-provided reference; not independently checked in this pass)
- Mistral changelog (retrieved from first-party public documentation on 2026-10-25; this is not a deployment test)
- mistral.ai — documentation/reference (supplier-provided reference; not independently checked in this pass)
Source documentation establishes vendor-published facts, not independently measured model quality, tenant access, contractual rights or current prices. Use a dated pricing ledger and an account-specific entitlement check for deployment decisions.